Concepts

Start here if you want to understand the system before writing YAML. Each page is 5–10 minutes; read top-to-bottom for the full conceptual tour, or jump to whichever piece you need.

Recommended order

1. What is an experience?
2. Code-as-Law — our central thesis
3. The 6 node kinds — at-a-glance + decision tree
4. State (SQLite blackboard)
5. Edges & control flow
6. Events & event log
7. Scheduler — Sequential vs Parallel
8. Dispatchers — the pluggable execution layer
9. Cache key + memoization — how oe resume works
10. Evolution loop — author → run → evolve closes :::

Core model

The mental model in one paragraph: an experience is a YAML-declared DAG of nodes running over a SQLite-backed state with a JSONL event log. The graph never changes at runtime — that's Code-as-Law. LLMs only fill in node bodies. After a run, the evolution loop can propose YAML diffs.

Page	What it covers
Experiences	The unit of authoring + the file layout
Code-as-Law	Why YAML, why not autonomous
Node kinds	The 6 kinds + the decision tree
State	SQLite blackboard + merge strategies
Control flow	Edges, when:, for_each, pipelines, loops
Events	The JSONL event log + 10 event types

Node kinds in depth

A page per kind — what's in the YAML, what the dispatcher does, the gotchas.

Page	Kind
tool	Deterministic JS function
agent	LLM with AJV-validated structured output
skill	Reusable SKILL.md package
dataset	File / SQLite / HTTP / MCP-resource loader
experience	Nested sub-experience
cli-agent	Subprocess to Claude Code / Codex / Gemini

Runtime

How the graph actually executes.

Page	What it covers
Scheduler	Sequential + Parallel, wave algorithm
Dispatchers	One per kind, two-phase resolve + run
Cache	Content-hash memoization for oe resume
Evolution loop	The advisor and the 3 operations

Where to next

• Want to write your first experience? → Guide: First Experience
• Want to see real examples? → Examples gallery
• Want to evaluate against other frameworks? → Compare
• Want the YAML schema? → Schema reference